Here are the results of E coli SpoT submission:
http://zhanglab.ccmb.med.umich.edu/I-TASSER/output/S181122/
As you can see 3 of the top 5 templates are in fact orthologs of SpoT, SpoT/RelA protein from S. equisimilis. Identity is 0.48 over 0.46 of sequence, consistent with orthologous relationship. However the resulting model 1 turns out looking like template 2, which has borderline identity - 0.1 but over 0.88 of the target sequence. In other words model doesn't look anything like the protein it is supposed to be. I had the same problem before, modeling topoisomerase and not getting anywhere close to any DNA-binding protein.